prores: fix multithreaded decoding case when slice quantisers are not the same

Since quantisation matrices are stored in context, decoding slices with
different quantisers in parallel leads to unpredictable content of
aforementioned matrices and wrong output picture thereof.
This commit is contained in:
Kostya Shishkov 2012-02-13 19:27:48 +01:00
parent dc4e57489f
commit 3ec623c22f

View File

@ -42,7 +42,10 @@ typedef struct {
int slice_num;
int x_pos, y_pos;
int slice_width;
int prev_slice_sf; ///< scalefactor of the previous decoded slice
DECLARE_ALIGNED(16, DCTELEM, blocks)[8 * 4 * 64];
DECLARE_ALIGNED(16, int16_t, qmat_luma_scaled)[64];
DECLARE_ALIGNED(16, int16_t, qmat_chroma_scaled)[64];
} ProresThreadData;
typedef struct {
@ -56,9 +59,6 @@ typedef struct {
uint8_t qmat_luma[64]; ///< dequantization matrix for luma
uint8_t qmat_chroma[64]; ///< dequantization matrix for chroma
int qmat_changed; ///< 1 - global quantization matrices changed
int prev_slice_sf; ///< scalefactor of the previous decoded slice
DECLARE_ALIGNED(16, int16_t, qmat_luma_scaled)[64];
DECLARE_ALIGNED(16, int16_t, qmat_chroma_scaled)[64];
int total_slices; ///< total number of slices in a picture
ProresThreadData *slice_data;
int pic_num;
@ -116,7 +116,6 @@ static av_cold int decode_init(AVCodecContext *avctx)
ctx->scantable_type = -1; // set scantable type to uninitialized
memset(ctx->qmat_luma, 4, 64);
memset(ctx->qmat_chroma, 4, 64);
ctx->prev_slice_sf = 0;
return 0;
}
@ -294,9 +293,11 @@ static int decode_picture_header(ProresContext *ctx, const uint8_t *buf,
for (i = 0; i < num_slices; i++) {
ctx->slice_data[i].index = data_ptr;
ctx->slice_data[i].prev_slice_sf = 0;
data_ptr += AV_RB16(index_ptr + i * 2);
}
ctx->slice_data[i].index = data_ptr;
ctx->slice_data[i].prev_slice_sf = 0;
if (data_ptr > buf + data_size) {
av_log(avctx, AV_LOG_ERROR, "out of slice data\n");
@ -561,11 +562,11 @@ static int decode_slice(AVCodecContext *avctx, void *tdata)
/* scale quantization matrixes according with slice's scale factor */
/* TODO: this can be SIMD-optimized a lot */
if (ctx->qmat_changed || sf != ctx->prev_slice_sf) {
ctx->prev_slice_sf = sf;
if (ctx->qmat_changed || sf != td->prev_slice_sf) {
td->prev_slice_sf = sf;
for (i = 0; i < 64; i++) {
ctx->qmat_luma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_luma[i] * sf;
ctx->qmat_chroma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_chroma[i] * sf;
td->qmat_luma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_luma[i] * sf;
td->qmat_chroma_scaled[ctx->dsp.idct_permutation[i]] = ctx->qmat_chroma[i] * sf;
}
}
@ -574,7 +575,7 @@ static int decode_slice(AVCodecContext *avctx, void *tdata)
(uint16_t*) (y_data + (mb_y_pos << 4) * y_linesize +
(mb_x_pos << 5)), y_linesize,
mbs_per_slice, 4, slice_width_factor + 2,
ctx->qmat_luma_scaled);
td->qmat_luma_scaled);
/* decode U chroma plane */
decode_slice_plane(ctx, td, buf + hdr_size + y_data_size, u_data_size,
@ -582,7 +583,7 @@ static int decode_slice(AVCodecContext *avctx, void *tdata)
(mb_x_pos << ctx->mb_chroma_factor)),
u_linesize, mbs_per_slice, ctx->num_chroma_blocks,
slice_width_factor + ctx->chroma_factor - 1,
ctx->qmat_chroma_scaled);
td->qmat_chroma_scaled);
/* decode V chroma plane */
decode_slice_plane(ctx, td, buf + hdr_size + y_data_size + u_data_size,
@ -591,7 +592,7 @@ static int decode_slice(AVCodecContext *avctx, void *tdata)
(mb_x_pos << ctx->mb_chroma_factor)),
v_linesize, mbs_per_slice, ctx->num_chroma_blocks,
slice_width_factor + ctx->chroma_factor - 1,
ctx->qmat_chroma_scaled);
td->qmat_chroma_scaled);
return 0;
}